Move all AtlasDB persisters to the new reusable type

[raiju]

Before this PR

Non-reusable persisters are created for each cell read/write. This causes a lot of object creation, which is particularly problematic when a new object mapper is created,

After this PR

==COMMIT_MSG== Enforce all AtlasDB persisters to be marked as reusable ==COMMIT_MSG==

Possible downsides?

• Should applying this fix be broken out into a separate excavator to ensure enough attention is paid to it? I haven't found a single non-reusable persister yet, but I'm sure they exist.
• @jeremyk-91 @Jolyon-S Are there reasons to not do this?

[changelog-app[bot]]

Generate changelog in changelog/@unreleased

Type

• [ ] Feature
• [x] Improvement
• [ ] Fix
• [ ] Break
• [ ] Deprecation
• [ ] Manual task
• [ ] Migration

Description Enforce all AtlasDB persisters to be marked as reusable

Check the box to generate changelog(s)

• [x] Generate changelog entry

[carterkozak]

Given that we own the persister API, I would prefer if we updated the API in such a way that it's difficult to get wrong rather than relying on static analysis.

[raiju]

I'd prefer that too; @jeremyk-91 is there a world where we could flip the expectations/always enforce this?

@carterkozak Would you be ok merging this with a plan to get persisters to be safe by default?

[jeremyk-91]

No objections to changing the API in the long term (in fact that's preferable, I remember this bit of defensiveness being somewhat annoying to add) - the main thing is how to get there. If this is likely to have people touch each of their persisters, I'm thinking if we should instead do:

• add @StatefulPersister annotation to AtlasDB (or something like that)
• require users to specify at least one
• (not 100% safe, but probably >99%) after some point in time, switch the default behaviour over

[raiju]

Refactored this to support the new setup (only ReusablePersisters are supported, Persisters are deprecated)